The Use of Bayesian Inference in the Design of an Endpoint Control System for the Basic Oxygen Steel Furnace

A digital simulation of the basic oxygen steel furnace was previously developed, and its output was compared with the available data taken from the literature. The output concentrations were within 10 percent of the literature data, while the simulated temperature was within 0.5 percent. The simulation is used as an off-line model of the process to design an endpoint control system which makes use of the available feedback from the process. Feedback consists of previously existing instrumentation for effluent gas analysis and an instrument designed for quick carbon analysis. The same instrument, independently conceived by Bethlehem Steel, has been proved effective by them. The control system uses Bayesian inference to evaluate process feedback optimally. Equations have been developed and a computational algorithm designed enabling real-time calculation of the probability of a carbon-temperature state given any control action and imperfect measurements. Because the objective function is almost symmetric and the cost of control is minimal compared to the value of an endpoint state, optimal control drives the expected state vector to the center of the tolerance region.